Writes Where's Your Ed At, a newsletter on the AI industry's finances, and hosts the Better Offline podcast covering the same territory in audio. Argues the AI boom rests on circular financing between a small group of companies and compute spending the underlying economics don't support, and tracks corporate debt, data center buildouts and venture capital flows as the evidence for that case.
Appears widely as a guest discussing the AI bubble, including on The Tech Report, On the Media, Prof G Markets, Big Technology and Newcomer Pod. Better Offline won a Webby Award and was named one of the best podcasts of 2024 by Esquire, Vulture and The Information.
The host asked why Alphabet's and Meta's earnings were punished when the market already knew they were spending heavily on AI. Zitron points to Google's search miss and argues its AI search mode earns less than the old blue-link results.
…think. So until things start really collapsing, the real world will not realize. But it will be when you are at work and you're suddenly told, okay, actually, don't use this as much. Oh, actually, please use these models. I've heard reports of this, but this is where you're really going to feel it.
Isaac Pound
…be spending this money on AI. We've known for a long time that they were going to spend this money on AI. They told us they were going to do it. And now we're seeing it on paper. So why is it having that impact now that when it's actually on paper and they're seeing it, it's dropping 10 to 12%?
Ed Zitron
…a selection of blue links to a 4,000-word essay that has a mushy peas recipe at the end. You now go and look on Google, but that is generating less revenue. It's quite literally what I said in 2023, amongst other early, early and wrong predictions around timing, by the way. Just was way ahead of my skis on the timing. Really, this did not end in 2025. But it's something I've seen, which is AI does not help a business get better. We do not have evidence for this. And indeed, we now have tangible evidence that it makes the business worse. That is crazy. It is crazy that Google spent $50 billion in capex in a quarter for a product that is making its business materially weaker. And what's crazy is it also shows that AI is propping it up mostly through Anthropic and OpenAI's compute spend. And the other thing we found out from the Anthropic IPO that was quite interesting was that apparently Anthropic considers 100% of the revenue that Google and Amazon get for selling their models through Bedrock and Vertex. So the two different platforms you plug in your models. What's…
Isaac Pound
Just finally, as a pretty stark figure in Slock's post, there's a little graph at the bottom, which kind of otherwise maybe would have gone unnoticed. It showed that 80% of the top 1% of token spend comes from venture capital. And that sounds very similar to the ramp figures that we got in…
Ed Zitron
It means that it is effectively impossible to justify the spend right now because it means, and the thing is, I think 64% could be low. I think it could be much higher. But nevertheless, even then, so 64% of your revenue is not real. It's demand that is only made possible when someone else is…
Isaac Pound
Wow. It's interesting. Thanks for taking the time. Thanks for having me. If you enjoyed today's episode and you want to hear more of the Tech Report, please consider liking and subscribing. Also, you can get episodes of the Tech Report wherever you get your podcasts.
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
The host cited a Bank of America survey in which fund managers named hyperscaler AI capex as their top worry, even as they keep investing in it. Zitron on why nobody moves first.
…the original guesses weren't based on anything. It's almost as if they never really thought about it. Because they didn't. That's the secret. None of them do. They are, they're just like, just like you and me. I don't know what I'm going to have it for breakfast in the morning, and neither do they.
Isaac Pound
…there until it all goes wrong or just before if they're smart enough. What it begs the question: from the perspective of the money manager, what is their way out of this if they both can't leave because they're expected to be in, but if they don't get out, then it's gonna, they're gonna lose. It's
Ed Zitron
…it is a problem in our financial system. It is something that journalism can actually push back on because unsafe things keep happening because of this. There is also the pure hypocrisy of it, which is that Bank of America, at the same time as publishing that is saying, yeah, there's an AI bubble, but you should keep investing in it. Because no one wants to be the person that leaves first. No one wants to be the person that runs for the door. Because they know that once they do that, everyone panics at once and blames them. However, doing that is the smartest call, but doing it too early will make you look stupid. Doing it too late will make you look stupid. So it's who is going to act, who is going to say, I'm full first. That's really what the question is. And it's impossible to tell. It's because it could be any of them at any time. But the fact they're all saying it is a pretty big worry. But honestly, it's all of them pointing guns at each other. Just like, I'm going to pull the trigger. You're going to pull it first. I'm going to shoot you in…
Isaac Pound
Well, Ezekron, thanks for taking the time. Thanks
Ed Zitron
for having me. If
Isaac Pound
you enjoyed today's episode and you want to hear more of the tech report, please consider liking and subscribing. Also, you can get episodes of the tech report wherever you get your podcasts.
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron on banks that talk up AI stocks in public while selling off the AI data center debt they bought months earlier. He's borrowing a line from Spider-Man: Into the Spider-Verse, and his point is that the debt markets, not stock prices, show where the risk is.
…I don't know. That worries me more than anything, but it's like, how worried do we be? We don't know. And that is actually deeply frustrating to me and scary. That's almost as scary as if it is as bad as we think. It's just like this big question mark for the biggest infrastructure thing ever.
Isaac Pound
…of the scale of the debt that they have is coming up against some sort of internal limits on industry exposure and things like that. But is that a reasonable assumption? And what's the ripple effect of buying and selling billions and billions of dollars of debt in such a small window of time?
Ed Zitron
So it is, like you said, it's very common for banks to take debt and then slice it up and sell it off to people. That's very common. Doing it so quickly, you would actually, in many cases, you would do it immediately. You would actually just, you would do a loan. The place to actually look is if they can't do that, if they are left sitting with it. That happened with Musk's Twitter deal when they couldn't sell the debt off. I think it also happened with the Paramount debt for buying Paramount with leathery Mr. Ellison and his big wet son David. I think that it's happening increasingly with data centers because they are trying to offload the risk, like you said, because they're over their risk limits. And actually, that is the biggest story here, which is how much they're selling off, interesting, could be big, isn't really clear. We don't really know how they're selling it, the reasons. What we do know is that every single bank has risk limits. There are limits to how much they can actually hold of any…
Isaac Pound
I just, one point of clarification, because we were talking about like they're selling it on because they're reaching sort of tolerance limits of this exposure. And then the people they sell it to will also have similar tolerance limits. I have no concept of what those limits are sort of in…
Ed Zitron
You're going to be horrified to hear it's probably the latter. Because so every time you hear an analyst jerking themselves from first principles about this stuff, they are doing so based on everything going perfectly. Especially if it's a big consensus holding. They are hearing, they hear all the…
Isaac Pound
Just finally, a Bank of America survey of global funding managers, which admittedly I did find through the Financial Times, they found that on average, the number one thing that these managers are worried about was not government debt. It was not private debt. It was hyperscalar capex on AI. And…
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron reacts to a Business Insider report that Amazon is taking GPUs months before it can install them, under what the report calls an early rack receipt program. He reads it as NVIDIA pushing chips onto customers, which would mean GPU sales overstate real demand for AI compute.
…industry is taking way more of these things than they currently need. What is scarce about that? Unless, of course, the vendor in question, NVIDIA, is saying, if you don't do this now, we will not sell you this in the future. And I think that's what's happening. It's classic Jensen Huang nonsense.
Isaac Pound
…a little bit. Which is that the I Told You So moments for you are coming a bit more frequently at the moment. And I'm obviously curious what you think that means in general, like what that increasing pace means. But I'm also curious how it feels to have the I Told You Sow moments coming quicker.
Ed Zitron
…sneaking in and get with a hair dryer. No, that's a joke. That already happened. But that's the thing. And people will get oity-toity, I'm sure, in the comments. And people responding to this will say, well, whoa, the hyperscalers do want these GPUs. They think they want them. They clearly have ordered too many, or at least they've ordered as many as they need, but way too early. I don't know. And I just, it shows that there is a fundamental disconnection between what people perceive as AI demand and what is actual AI demand. And this is a problem across the industry. It's a problem that I've been analyzing for three years, which is we cannot actually measure the true demand for AI. Because even when it comes to, and I know we're going to get into this as well, that ramp stat of 8% of OpenAI and Anthropics revenues, 1% of their customers. Except most of those customers are venture-backed AI startups who are selling…
Isaac Pound
If we go back to the rate at which Amazon is buying these racks, these GPUs, a delay of a few weeks would mean that the stack of warehouse GPUs would presumably then grow by billions of dollars. A four-week delay, that's another $3 to $5 billion you have waiting.
Ed Zitron
Yeah, exactly.
Isaac Pound
But I've seen in places, and I'm sure you have as well, that wait time for energy components is going up exponentially, let alone the construction itself. There's loads of these things, which are just there's the moratorium. There's lots of things that causing delays, which are a little bit longer…
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Asked whether CoreWeave has reached a tipping point, Zitron says it will not blow up all at once. With its debt trading at double-digit yields, he expects missed payments and failed fundraising, and he calls it a "disease-ridden company" rather than one about to explode.
Which sums it up really nicely. And I was just wondering, like, how close you think we are to a bit of a tipping point with Coreweave?
Ed Zitron
…they buy things, whatever, but they're not holding quite as much. Why? It's probably because they don't think they're going to get paid long term. And to quote the CEO of the company from Margin Core, they're selling to willing buyers. But the thing is, it's not so much Core Weave and the AI bubble writ large. One of the mistakes I've made in my analyses is the early ones, I would say, is that I've been looking for an explosion when you're looking for, to quote my mate, Casey, a dead body. It's a slow boil. You find corpses here and there. Things die. Things don't happen. Contracts don't get signed. Debt doesn't get raised. So it's less that Core Weave suddenly explodes and more Coreweave misses payments and Core Weave cannot raise more debt. There was an analyst Isaac Pound that said that Core Weave needs to raise another 100 billion or so in debt by 2030. So across the next three years, at what? Like you're talking 10, 11, 12, 13% rates. At some point, even if you think AI is the most important thing ever, there…
Isaac Pound
for it. Go for it. Eager to. Go for it.
Ed Zitron
Okay. So we should launch into the Anthropic. So the reason that the markets are puking today, which is Thursday, so this will be running on Friday, is because Anthropic's Q3 numbers came out in a bleak from an investor meeting leading up to their IPO. Now, their Q3 numbers were they had $16.8…
Isaac Pound
They're making so much money off people stealing from us.
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron was asked how a data-center company paying 11.5% to 13.5% on its own debt gets investment-grade pricing when a big tech company is the customer. His complaint is with the credit agencies: they rate the debt on the tenant who will eventually pay rent, and skip over whether the building, which doesn't exist yet, will get built.
…bubble. It's nothing like it. Also, that was telecommunications companies rather than hyperscalers. The actual software companies and hardware companies really weren't the ones other than a few here and there. So, and also the debt wasn't as bad. It's just so strange. It's just a very strange era.
Will Guyatt
Core weave is paying 11.5% to 13.5% on its own debt, but it got investment-grade pricing when Meta was the customer. And we touched on this a while back. Is borrowing on a hyperscaler's name like an escape hatch, or is it just moving the risk onto those companies?
Ed Zitron
…at the end. It is, I guess, they also couldn't find any comparison to it. But it's why there are all of these projects with these, like Coreweave has a few now. I think maybe I forget one of the other NeoClouds has an investment-grade vehicle because of a connection to a hyperscaler. And it's obscene because it's not, if you tried to underwrite a consumer loan, let's say, in fact, great example. If you tried to get a mortgage and your parents co-signed the loan and they said, we will be the ones that make the payments because we are going to live in the second bedroom. The bank would start asking real questions about your parents. They would say, I don't know. This doesn't seem like a mortgage with you. It seems like a mortgage with your parents, except the credit risk is with you. I don't like that. And then they'd start turning their nose up at it. But because it's a data center, they're like, oh, brilliant. Yeah. In fact, it's actually sillier than that. It's you are doing a house that you're building that your parents…
Will Guyatt
So, Amadai said that public markets are well suited to capital-intensive
Ed Zitron
AI. And your answer has basically been, really, are they? Yeah. Why isn't going public the funding fix that people seem to assume is going to be? So, it gets back to that junk debt thing I was talking about. So, let's start with SpaceX and Core Weave. People will say, Well, how did SpaceX get…
Will Guyatt
Who stops them there going public?
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron and his guest had been speculating about where the GPUs bought for AI data centers physically are, given how much capacity is not yet powered. "They" are fellow tech journalists, who he says shrug when he raises it. The dollar figure is his own estimate.
Yeah, are they in a data center but without power? That wouldn't make sense, though, because if the data center didn't have power, that would actually be dangerous for them.
Speaker 2
…warm shells, which to me, in your situation there, I think the only way that happens... while his warm shell statement is true, is if it's a data center that doesn't have power yet. And I just don't see that happening, you know, like, cause then you're probably, you're not even permitted. Right.
Speaker 1
So it's just, and again, you go to the various journalists about this and you try and make them talk about it and they go, ah, you know, it really is. This is actually what jokifies me. It's when you try and talk to your peers and they're like, ah, yeah, you know, it's bad, I guess, but who's to say. And it's just like, okay, but we've sold possibly 200, $ 300 billion worth of GPUs that have gone nowhere and they're planning to buy more. This alone is a story. And they look at you like you're reading algebra to a dog. Yeah.
Speaker 2
I mean, it's like, I feel like a product physically exists probably.
Speaker 1
I hope so. Legally speaking. Like if it doesn't, and this is just literally balance sheet transfers, that's so bad. Right.
Speaker 2
I mean, that's how I feel too. It's like, I think there's a certain level of awareness and I don't, everybody is, people aren't stupid, I think. And it seems unlikely that it would be a pure balance sheet type of.
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron's argument, built from the "construction in progress" lines in hyperscaler earnings reports, is that a large share of the GPUs Nvidia has sold are sitting in warehouses or in data centers without power. He is explicit that it's his own estimate from incomplete disclosure — Microsoft doesn't report the figure at all. The point he draws from it is that chip sales measure willingness to speculate on AI, not demand for it.
…the data centers, Larry? How does it work? It's crazy that this is not all over the finance press and the business press and the tech press. I mean, they're covering the sporce majeure thing, but the brittle state of Oracle and its relationship with OpenAI is going to be the story of the AI bubble.
Isaac Pound
…this week into the numbers behind how many GPUs there are sitting around. We did a little bit of a React episode, kind of, I guess it was to the story with Microsoft having not brought on as much compute as we thought they did. Since then, what have you found and how many GPUs are being hoarded?
Ed Zitron
…installed one. Yeah, they installed a server in a building. And when you go and look, it's in the middle of a floor, which is not how data centers look. So it's like, yeah, you plug one of these things in, but it's the key jingling. It's a jingle jingle. Oh, look, we've got a Vera Rubin. Oh, it's so good and fast. And everyone goes, yay, everything's happening. Except nothing's happening. Like, it sounds like maybe $150 billion worth of GPUs out of the 535 billion, maybe 500 billion sold, have actually been installed anywhere. And this is bad for NVIDIA. It's bad for Broadcom. Broadcom estimates they're getting, sorry, analysts estimate that Broadcom's going to make about $300 billion selling TPUs to Anthropic and Google and also chips to OpenAI in the next three years. How's that going to, first of all, how are they going to pay for them, of…
Isaac Pound
I mean, one of the big questions on the topic of big questions that we came away with last week was: if why are people still buying more GPUs if they're just putting them in a warehouse? Have you gained any more insight on that at all?
Ed Zitron
I think that it's a combination of things. One, I think the hyperscalers are doing just dim-witted industrial planning. They are saying, okay, I'm going to build expazillion gigawatts of G of AI compute. I need to buy this many GPUs. I also think that Jensen Huang of NVIDIA has made it very clear…
Isaac Pound
Where does it leave? I mean, GPU bag debt. If I cannot believe that it was sold on the terms or the knowledge or the expectation that there's all these chips just sitting around not being used. And the thing with the depreciation that I kind of get stuck on, which does lead, does connect to this as…
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron had been attacking an ex-Anthropic researcher's public warning about AI risk when his co-host pushed back, asking whether models conducting cyberattacks are a real concern. He says it is, then turns the industry's stock question about AI falling into the wrong hands back onto the two labs themselves. The rest of his answer argues the danger is unreliable software handed too much autonomy, not machine intent.
…of that happening. It's just very frustrating. It boils my blood because modern journalism is meant to be full of smart people. These people went to Ivy League degree. They've got Ivy League degrees. They went to Oxford and Cambridge. And yet, they get hit by a parked car. They fall for anything.
Isaac Pound
What about the fact that various models have conducted cyber attacks on external platforms? I mean, poorly configured sandboxes aside, of course. Is the fact that it's kind of out of alignment and that kind of stuff not a concern at all?
Ed Zitron
It is a concern. I want to be abundantly clear about something. These companies, these people keep saying, what if AI falls into the wrong hands? Dangerous AI is already in the wrong hands. Anthropic and open AI. The hugging face attack, and Cal Newport's made some great points about this, is a result of software doing what it was trained to do, but the software itself is inconsistent and hard to control. Now, out of control does not mean, oh, it's doing stuff without us. Oh, it's intelligent. It means that you have built a technology on top of large language models, a probabilistic technology that is…
Isaac Pound
and
Ed Zitron
being able to use as much as it could, it was able to just brute force through. And the thing is, if a human, if a regular person did exactly this, if a regular person had, I don't know, they bought, I don't know how they did afford 10,000 GPUs to rent and they hacked Hugging Face, they would be in…
Isaac Pound
The thing that sticks out for me is the timing. Maybe not necessarily of him quitting now, but also just generally this open AI saying we're going to slow down development. Anthropic. CEO and various other thousand people or so saying we need to pace development while each launch is becoming more…
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
Zitron had spent the segment arguing that a thin slice of heavy customers carries almost all AI revenue, and that boosters answer that argument by demanding you admit ChatGPT is better. This is his objection to the demand itself: it treats a paid software product as a position on right and wrong. He goes on to say they want the argument kept there so nobody looks at who is actually paying.
…fact that Broadcom's largest customers are going to be Anthropic and OpenAI, that's also bad. So we just have all of these consequences waiting around the corner and everyone just saying, it's not a big deal. We'll deal with it. Like, stop thinking. Stop asking questions. Nothing bad ever happens.
Isaac Pound
I mean, so if we take that 80% from 1% statistic and we kind of extrapolate it out, what does it mean for the narrative that AI is this inevitable thing that has such broad demand and such broad use cases that you got to get on board or get in the bin?
Ed Zitron
…a. A ridiculous thing right now where AI boosters will come to you and they will talk to you like a religious fanatic and they will say, Admit it! Admit that ChatGPT is better! Admit to me, you must concede that ChatGPT is better. First of all, none of these idiots will concede a single damn thing about the finances, so screw them. Second of all, this is software you pay for. This is not morals. It is not a moral good to be correct about how much AI can do. On top of that, they're not correct on it, but it's not like liking or disliking a software product you pay money for is not a ideological thing. It's stupid to see it that way. And they want it to be like that so they can push this through so that you'll all stop thinking about one thing, which is that most people will not pay for AI. Even though it's got better in the last year, most people…
Isaac Pound
You know that feeling. Too many updates? Too many meetings? Too many things falling through the cracks. Monday.com was built for that gap. The AI work platform designed from the ground up for people and agents to work side by side and deliver more together. Drafting the brief, flagging risks…
Speaker 2
A BetterHelp ad. Matching with my therapist was very quick and easy. I just filled out a couple of questions. Before you know it, I was matched with a therapist. I just felt like a weight just was lifted off of my shoulders. I felt like it was a good match from the first time we talked. ButterHelp…
Speaker 1
Pay testimonials. Results may vary.
Speaker names come from our own diarization. The position is estimated from where the line sits in the episode.
